You are viewing a plain text version of this content. The canonical link for it is here.
Posted to notifications@kyuubi.apache.org by "yaooqinn (via GitHub)" <gi...@apache.org> on 2023/04/10 08:36:30 UTC
[GitHub] [kyuubi] yaooqinn commented on a diff in pull request #4615: Support stage level schedule for final write stage
yaooqinn commented on code in PR #4615:
URL: https://github.com/apache/kyuubi/pull/4615#discussion_r1161545054
##########
docs/extensions/engines/spark/rules.md:
##########
@@ -63,26 +63,31 @@ Now, you can enjoy the Kyuubi SQL Extension.
Kyuubi provides some configs to make these feature easy to use.
-| Name | Default Value | Description | Since |
-|---------------------------------------------------------------------|---------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-------|
-| spark.sql.optimizer.insertRepartitionBeforeWrite.enabled | true | Add repartition node at the top of query plan. An approach of merging small files. | 1.2.0 |
-| spark.sql.optimizer.insertRepartitionNum | none | The partition number if `spark.sql.optimizer.insertRepartitionBeforeWrite.enabled` is enabled. If AQE is disabled, the default value is `spark.sql.shuffle.partitions`. If AQE is enabled, the default value is none that means depend on AQE. | 1.2.0 |
-| spark.sql.optimizer.dynamicPartitionInsertionRepartitionNum | 100 | The partition number of each dynamic partition if `spark.sql.optimizer.insertRepartitionBeforeWrite.enabled` is enabled. We will repartition by dynamic partition columns to reduce the small file but that can cause data skew. This config is to extend the partition of dynamic partition column to avoid skew but may generate some small files. | 1.2.0 |
-| spark.sql.optimizer.forceShuffleBeforeJoin.enabled | false | Ensure shuffle node exists before shuffled join (shj and smj) to make AQE `OptimizeSkewedJoin` works (complex scenario join, multi table join). | 1.2.0 |
-| spark.sql.optimizer.finalStageConfigIsolation.enabled | false | If true, the final stage support use different config with previous stage. The prefix of final stage config key should be `spark.sql.finalStage.`. For example, the raw spark config: `spark.sql.adaptive.advisoryPartitionSizeInBytes`, then the final stage config should be: `spark.sql.finalStage.adaptive.advisoryPartitionSizeInBytes`. | 1.2.0 |
-| spark.sql.analyzer.classification.enabled | false | When true, allows Kyuubi engine to judge this SQL's classification and set `spark.sql.analyzer.classification` back into sessionConf. Through this configuration item, Spark can optimizing configuration dynamic. | 1.4.0 |
-| spark.sql.optimizer.insertZorderBeforeWriting.enabled | true | When true, we will follow target table properties to insert zorder or not. The key properties are: 1) `kyuubi.zorder.enabled`: if this property is true, we will insert zorder before writing data. 2) `kyuubi.zorder.cols`: string split by comma, we will zorder by these cols. | 1.4.0 |
-| spark.sql.optimizer.zorderGlobalSort.enabled | true | When true, we do a global sort using zorder. Note that, it can cause data skew issue if the zorder columns have less cardinality. When false, we only do local sort using zorder. | 1.4.0 |
-| spark.sql.watchdog.maxPartitions | none | Set the max partition number when spark scans a data source. Enable MaxPartitionStrategy by specifying this configuration. Add maxPartitions Strategy to avoid scan excessive partitions on partitioned table, it's optional that works with defined | 1.4.0 |
-| spark.sql.optimizer.dropIgnoreNonExistent | false | When true, do not report an error if DROP DATABASE/TABLE/VIEW/FUNCTION/PARTITION specifies a non-existent database/table/view/function/partition | 1.5.0 |
-| spark.sql.optimizer.rebalanceBeforeZorder.enabled | false | When true, we do a rebalance before zorder in case data skew. Note that, if the insertion is dynamic partition we will use the partition columns to rebalance. Note that, this config only affects with Spark 3.3.x. | 1.6.0 |
-| spark.sql.optimizer.rebalanceZorderColumns.enabled | false | When true and `spark.sql.optimizer.rebalanceBeforeZorder.enabled` is true, we do rebalance before Z-Order. If it's dynamic partition insert, the rebalance expression will include both partition columns and Z-Order columns. Note that, this config only affects with Spark 3.3.x. | 1.6.0 |
-| spark.sql.optimizer.twoPhaseRebalanceBeforeZorder.enabled | false | When true and `spark.sql.optimizer.rebalanceBeforeZorder.enabled` is true, we do two phase rebalance before Z-Order for the dynamic partition write. The first phase rebalance using dynamic partition column; The second phase rebalance using dynamic partition column Z-Order columns. Note that, this config only affects with Spark 3.3.x. | 1.6.0 |
-| spark.sql.optimizer.zorderUsingOriginalOrdering.enabled | false | When true and `spark.sql.optimizer.rebalanceBeforeZorder.enabled` is true, we do sort by the original ordering i.e. lexicographical order. Note that, this config only affects with Spark 3.3.x. | 1.6.0 |
-| spark.sql.optimizer.inferRebalanceAndSortOrders.enabled | false | When ture, infer columns for rebalance and sort orders from original query, e.g. the join keys from join. It can avoid compression ratio regression. | 1.7.0 |
-| spark.sql.optimizer.inferRebalanceAndSortOrdersMaxColumns | 3 | The max columns of inferred columns. | 1.7.0 |
-| spark.sql.optimizer.insertRepartitionBeforeWriteIfNoShuffle.enabled | false | When true, add repartition even if the original plan does not have shuffle. | 1.7.0 |
-| spark.sql.optimizer.finalStageConfigIsolationWriteOnly.enabled | true | When true, only enable final stage isolation for writing. | 1.7.0 |
-| spark.sql.finalWriteStage.eagerlyKillExecutors.enabled | false | When true, eagerly kill redundant executors before running final write stage. | 1.8.0 |
-| spark.sql.finalWriteStage.retainExecutorsFactor | 1.2 | If the target executors * factor < active executors, and target executors * factor > min executors, then inject kill executors or inject custom resource profile. | 1.8.0 |
+| Name | Default Value | Description | Since |
+|---------------------------------------------------------------------|----------------------------------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-------|
+| spark.sql.optimizer.insertRepartitionBeforeWrite.enabled | true | Add repartition node at the top of query plan. An approach of merging small files. | 1.2.0 |
+| spark.sql.optimizer.insertRepartitionNum | none | The partition number if `spark.sql.optimizer.insertRepartitionBeforeWrite.enabled` is enabled. If AQE is disabled, the default value is `spark.sql.shuffle.partitions`. If AQE is enabled, the default value is none that means depend on AQE. | 1.2.0 |
+| spark.sql.optimizer.dynamicPartitionInsertionRepartitionNum | 100 | The partition number of each dynamic partition if `spark.sql.optimizer.insertRepartitionBeforeWrite.enabled` is enabled. We will repartition by dynamic partition columns to reduce the small file but that can cause data skew. This config is to extend the partition of dynamic partition column to avoid skew but may generate some small files. | 1.2.0 |
+| spark.sql.optimizer.forceShuffleBeforeJoin.enabled | false | Ensure shuffle node exists before shuffled join (shj and smj) to make AQE `OptimizeSkewedJoin` works (complex scenario join, multi table join). | 1.2.0 |
+| spark.sql.optimizer.finalStageConfigIsolation.enabled | false | If true, the final stage support use different config with previous stage. The prefix of final stage config key should be `spark.sql.finalStage.`. For example, the raw spark config: `spark.sql.adaptive.advisoryPartitionSizeInBytes`, then the final stage config should be: `spark.sql.finalStage.adaptive.advisoryPartitionSizeInBytes`. | 1.2.0 |
+| spark.sql.analyzer.classification.enabled | false | When true, allows Kyuubi engine to judge this SQL's classification and set `spark.sql.analyzer.classification` back into sessionConf. Through this configuration item, Spark can optimizing configuration dynamic. | 1.4.0 |
+| spark.sql.optimizer.insertZorderBeforeWriting.enabled | true | When true, we will follow target table properties to insert zorder or not. The key properties are: 1) `kyuubi.zorder.enabled`: if this property is true, we will insert zorder before writing data. 2) `kyuubi.zorder.cols`: string split by comma, we will zorder by these cols. | 1.4.0 |
+| spark.sql.optimizer.zorderGlobalSort.enabled | true | When true, we do a global sort using zorder. Note that, it can cause data skew issue if the zorder columns have less cardinality. When false, we only do local sort using zorder. | 1.4.0 |
+| spark.sql.watchdog.maxPartitions | none | Set the max partition number when spark scans a data source. Enable MaxPartitionStrategy by specifying this configuration. Add maxPartitions Strategy to avoid scan excessive partitions on partitioned table, it's optional that works with defined | 1.4.0 |
+| spark.sql.optimizer.dropIgnoreNonExistent | false | When true, do not report an error if DROP DATABASE/TABLE/VIEW/FUNCTION/PARTITION specifies a non-existent database/table/view/function/partition | 1.5.0 |
+| spark.sql.optimizer.rebalanceBeforeZorder.enabled | false | When true, we do a rebalance before zorder in case data skew. Note that, if the insertion is dynamic partition we will use the partition columns to rebalance. Note that, this config only affects with Spark 3.3.x. | 1.6.0 |
+| spark.sql.optimizer.rebalanceZorderColumns.enabled | false | When true and `spark.sql.optimizer.rebalanceBeforeZorder.enabled` is true, we do rebalance before Z-Order. If it's dynamic partition insert, the rebalance expression will include both partition columns and Z-Order columns. Note that, this config only affects with Spark 3.3.x. | 1.6.0 |
+| spark.sql.optimizer.twoPhaseRebalanceBeforeZorder.enabled | false | When true and `spark.sql.optimizer.rebalanceBeforeZorder.enabled` is true, we do two phase rebalance before Z-Order for the dynamic partition write. The first phase rebalance using dynamic partition column; The second phase rebalance using dynamic partition column Z-Order columns. Note that, this config only affects with Spark 3.3.x. | 1.6.0 |
+| spark.sql.optimizer.zorderUsingOriginalOrdering.enabled | false | When true and `spark.sql.optimizer.rebalanceBeforeZorder.enabled` is true, we do sort by the original ordering i.e. lexicographical order. Note that, this config only affects with Spark 3.3.x. | 1.6.0 |
+| spark.sql.optimizer.inferRebalanceAndSortOrders.enabled | false | When ture, infer columns for rebalance and sort orders from original query, e.g. the join keys from join. It can avoid compression ratio regression. | 1.7.0 |
+| spark.sql.optimizer.inferRebalanceAndSortOrdersMaxColumns | 3 | The max columns of inferred columns. | 1.7.0 |
+| spark.sql.optimizer.insertRepartitionBeforeWriteIfNoShuffle.enabled | false | When true, add repartition even if the original plan does not have shuffle. | 1.7.0 |
+| spark.sql.optimizer.finalStageConfigIsolationWriteOnly.enabled | true | When true, only enable final stage isolation for writing. | 1.7.0 |
+| spark.sql.finalWriteStage.eagerlyKillExecutors.enabled | false | When true, eagerly kill redundant executors before running final write stage. | 1.8.0 |
+| spark.sql.finalWriteStage.retainExecutorsFactor | 1.2 | If the target executors * factor < active executors, and target executors * factor > min executors, then inject kill executors or inject custom resource profile. | 1.8.0 |
+| spark.sql.finalWriteStage.resourceIsolation.enabled | false | When true, make final write stage resource isolation using custom RDD resource profile. | 1.2.0 |
Review Comment:
please make a follow-up to fix the doc 1.2.0 -> 1.8.0
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: notifications-unsubscribe@kyuubi.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: notifications-unsubscribe@kyuubi.apache.org
For additional commands, e-mail: notifications-help@kyuubi.apache.org